GB2471536A

GB2471536A - Non-intrusive utility monitoring

Info

Publication number: GB2471536A
Application number: GB1000695A
Authority: GB
Inventors: James Donaldson; Sarah Surrall; Alex Matthews; Semen Trygubenko; Malcolm Mcculloch
Original assignee: Intelligent Sustainable Energy Ltd
Current assignee: Intelligent Sustainable Energy Ltd
Priority date: 2010-01-15
Filing date: 2010-01-15
Publication date: 2011-01-05
Anticipated expiration: 2030-01-15
Also published as: GB201000695D0; GB2471536B

Abstract

A non-intrusive method of monitoring the use of a utility 10 (e.g. electricity, water, gas or oil) supplied to multiple appliances 12 and using the monitored value to detect the operation of and/or identify — an active appliance. The method comprises: receiving values representative of the total use of the utility by the appliances; analysing the received values using multiple analysis modules (57, Fig. 2), each of which corresponds to a respective type of utility usage and is arranged to calculate a respective confidence value indicative of a confidence that the respective type of utility usage has occurred; and performing a fuzzy logic analysis of the calculated confidence values so as to identify the operation of an appliance. The type of utility usage may be usage representative of a resistive device with a constant steady-state load; a predominantly resistive device employing intra-cycle switching to variably control the power supply to a load; or an induction motor.

Description

NON-INTRUSIVE UTILITY MONITORING

FIELD OF THE INVENTION

The present invention relates to a method and apparatus for Non-Intrusive Utility Monitoring for monitoring the use of at least one utility supplied to a plurality of appliances.

BACKGROUND OF THE INVENTION

Non-Intrusive Load Monitoring (NILM), or Non-Intrusive Appliance Load Monitoring (NIALM) is, in essence, a method of analysing the voltage and current going into a house and thereby deducing which household appliances are being used at a particular time, as well as the individual energy consumption of each appliance.

Many NILM implementations work by tracking the steady state power drawn by the load and measuring significant changes which correspond to appliances turning on and off. For example, following the turn on of a 100W incandescent light, one would observe a change in real power of around 100W and minimal changes in the reactive power and the harmonic spectrum. A NILM would register these changes and search through its database to see whether this appliance power matched those of a previously observed appliance. If so, then it would infer that the appliance had turned on. If not, then it would infer that a previously unknown appliance had turned on for the first time. Further, by analysing characteristics of the signature, it would be possible to not only measure the energy consumption of the appliance, but to also identify the type of appliance (e.g. Kettle or Toaster).

To date, the majority of NILM systems have worked by taking low frequency measurements (of the order of Hz) of various power line features (power, reactive power and harmonic spectra) and have attempted to identify appliances from these features.

An overview of NILM technology can be obtained from US 4858141 (Hart).

In US 4858141, a non-intrusive monitor of energy consumption of residential appliances is described in which sensors, coupled to the power circuits entering a residence, supply analogue voltage and current signals which are converted to digital format and processed to detect changes in certain residential load parameters, i.e. admittance. Cluster analysis techniques are employed to group change measurements into certain categories, and logic is applied to identify individual appliances and the energy consumed by each.

The present invention seeks to provide an alternative NILM system which provides various advantages over those of the prior art.

SUMMARY OF THE INVENTION

io According to a first aspect of the present invention, there is provided a method of non-intrusive utility monitoring for monitoring the use of at least one utility supplied to a plurality of appliances. The method comprises: receiving utility values representative of the total use of the at least one utility by the plurality of appliances; analysing the received utility values using a plurality of analysis modules, wherein each analysis module corresponds to a respective predetermined type of utility usage, and wherein each analysis module is arranged to calculate, based on the received utility values, a respective confidence value indicative of a confidence that the respective predetermined type of utility usage has occurred; and performing a fuzzy logic analysis of the calculated confidence values so as to identify the operation of an appliance.

Advantageously, the predetermined type of utility usage for one of the plurality of analysis modules is usage representative of a resistive device with a relatively constant steady-state load, such as a resistive heater.

Advantageously, the predetermined type of utility usage for one of the plurality of analysis modules is usage representative of a predominantly resistive device employing intra-cycle switching to variably control the power supplied to a load (e.g. a TRIAC-controlled dimmer switch). The TRIAC is just one specific example of a controlled switch for such devices; other examples include: SCRs (silicon-controlled rectifiers), thyristors and transistors.

Advantageously, the predetermined type of utility usage for one of the plurality of analysis modules is usage representative of an induction motor wherein a path traced by real power values against corresponding reactive power values over a time period of interest comprises one or more substantially circular arcs.

Advantageously, the at least one utility comprises electricity, and the utility values comprise values representative of the electrical current and/or the electrical voltage supplied to the plurality of appliances.

Advantageously, the at least one utility comprises one or more of electricity, water, gas and oil.

Advantageously, the method further comprises detecting at least one utility event based on the received utility values.

Advantageously, each confidence value is a degree of membership of a respective membership function corresponding to the respective predetermined type of utility usage.

According to a second aspect of the present invention, there is provided a computer program comprising computer-executable code that when executed on a computer system, causes the computer system to perform a method according to the first aspect.

According to a third aspect of the present invention, there is provided a computer-readable medium storing a computer program according to the second aspect.

According to a fourth aspect of the present invention, there is provided a computer program product comprising a signal comprising a computer program according to the second aspect.

According to a fifth aspect of the present invention, there is provided a non-intrusive utility monitoring apparatus for monitoring the use of at least one utility supplied to a plurality of appliances. The apparatus comprises: an input section arranged to receive utility values representative of the total use of the at least one utility by the plurality of appliances; a plurality of analysis modules, wherein each analysis module corresponds to a respective type of utility usage, and wherein each analysis module is arranged to analyse the received utility values so as to calculate, based on the received utility values, a respective confidence value indicative of a confidence that the respective type of utility usage has occurred; and a fuzzy logic module arranged to perform a fuzzy logic analysis of the calculated confidence values so as to identify the operation of an appliance.

Advantageously, the apparatus further comprises a processor which comprises the plurality of analysis modules and the fuzzy logic module. Thus, the modules may be separate components/processors arranged to perform the separate functions, or may be a single componentiprocessor arranged to perform a combination of functions together (e.g. where it is difficult to separate the functionality of the various modules).

Other preferred features of the present invention are set out in the appended claims.

BRIEF DESCRIPTION OF THE DRAWINGS

Embodiments of the present invention will now be described by way of example with reference to the accompanying drawings in which: Figure 1 depicts schematically a Non-Intrusive Utility Monitoring (NIUM) system using a NIUM apparatus according to one embodiment of the invention; Figure 2 depicts schematically the NIUM apparatus of Figure 1; Figure 3 shows the Fourier domain representation of the current waveform from a single appliance; Figure 4 shows a typical current waveform for a TRIAC-type device; Figure 5 shows an example of a membership function for a resistive analysis module of the NIUM apparatus of Figures 1 and 2; and Figure 6 depicts schematically one embodiment of a fuzzy logic module of the NIUM apparatus of Figures 1 and 2

DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT

An overview of the present Non-Intrusive Utility Monitoring (NIUM) system is shown in Figure 1.

In Figure 1, the electricity supply to a site (e.g. a house, apartment, office, shop, school, building, etc.) is denoted IOA. The electricity is supplied to one or more of a plurality of appliances 12A, 12B, 12C, 12... by means of conventional wiring 14. The appliances and wiring are simply shown schematically in Figure 1, but may, of course, be configured in any appropriate way, such as via a consumer unit with circuit breakers or fuses, and with one or more ring main circuits with branches or spurs. An electricity meter 16A is provided to measure the total instantaneous current being provided to all of the appliances 12 from the supply 10, and also to measure the instantaneous voltage of the electricity supply 10. The current is measured by any suitable sensor, for example a current clamp placed around one of the conductors of the electricity supply wiring 14. The current clamp typically comprises a magnetizable material, such as ferrite, which forms a magnetic circuit around the conductor, and acts as a transformer to induce a voltage in a secondary winding around the magnetizable material, from which the current flowing in the supply wiring 14 can be obtained. As an alternative to this current-transformer, a Hall-effect sensor can be used to measure the magnetic field in the loop of magnetizable material around the wire which is related to the current flowing through the wire. Other suitable ways may, is of course, be used for sensing the current.

The voltage of the electricity supply can also be measured by any suitable volt meter. This, of course, typically requires access to two of the conductors in the wiring 14. This can be achieved, for example, by probes which strap around the respective cables and have spikes which penetrate the insulation to make contact with the conductor. Alternatively, connections could be made to terminals in the consumer unit, or, for example, at a location where fuses or circuit breakers are insertable. Non-invasive capacitive voltage detectors could also be used.

In Figure 1, some appliances 12 (e.g. washing machine or power-shower) are also connected to the water supply lOB. Alternatively, some appliances (e.g. kitchen sink tap) may only be connected to the water supply lOB and not the electricity supply bA. A water meter 16B detects the flow of water. Some appliances 12 may additionally/alternatively be connected to the supply of other utilities bC, 1OD Corresponding utility meters 16C, 16D, ... are provided to detect the overall utility usage of each utility 1 OC, 1 OD by the appliances 12 atthesitell.

As shown in Figure 1, the utility meters 16 are connected to an NIUM apparatus 20. It is, of course, possible that some or all of the utility meters 16 are incorporated within the apparatus 20, for example that wires connect the supply wiring 14 to the apparatus 20, and the voltage is measured within the apparatus 20. Alternatively, in a different embodiment, the one or more of the utility meters 16 may be self-contained and may communicate with the apparatus 20 wirelessly, e.g. by sending analogue or digital values of the instantaneous current and instantaneous voltage. In one option, the apparatus 20 can derive its own power supply by virtue of being connected to the portion of the electricity meter 16A for measuring voltage. In one particular form of this, the apparatus 20 is simply plugged into an electrical outlet in the same way as an appliance 12 to obtain its power supply and also to measure the supply voltage. However, in the preferred embodiment, the apparatus 20 and utility meters 16 are conveniently located near where the utility supplies 10 enters the building 11, such as near where the conventional electricity meter is or would be located.

The apparatus 20 comprises a number of different units, namely an input section 22, a clock 24, a processor 26, a store or memory 28, and an output section 40. It is possible to implement each of the various units as dedicated hard-wired electronic circuits; however the various units do not have to be separate from each other, and could all be integrated onto a single electronic chip such as an Application Specific Integrated Circuit (ASIC) or Field Programmable Gate Array (FPGA) or Digital Signal Processor (DSP) device. Furthermore, the units can be embodied as a combination of hardware and software, and the software can be executed by any suitable general-purpose microprocessor, such that in one embodiment the apparatus 20 could be a conventional personal computer (PC). The software would take the form of one or more computer programs having computer instructions which, when executed by a computer (e.g. processor 26) carry out a method according to an embodiment of the present invention as discussed below. The computer programs may be stored on a computer-readable storage medium, such as a magnetic disc, optical disc (e.g. a CD or DVD), etc. The input section 22 of the apparatus 20 receives current and voltage values from the electricity meter 16A. The values are input or measured preferably multiple times per cycle of the alternating electricity supply to a level of accuracy as required by the application. If the values are supplied as analogue voltages, then the input section 22 may comprise, for example, an analogue to digital converter, such that the rest of the apparatus 20 can be implemented using digital electronics. The input section 22 of the apparatus 20 also receives values representative of use of water (e.g. water flow rate measurements or water pressure measurements) from the water meter 16B. Similarly, other values may be provided to the input section 22 by the other utility meters 16C, 16D, (e.g. other utility flow rate measurements such as oil or gas flow rate measurements, or other utility pressure measurements such as oil or gas pressure measurements). The input section 22 also receives time data from the clock 24 which provides the actual present time. The clock 24 could, of course, be integral with other components of the apparatus 20, or the apparatus 20 could receive a clock signal fromanexternal source such as a transmitter broadcasting time data. In one preferred embodiment the clock 24 comprises a quartz is oscillator together with other timer circuitry that is an integral part of the processor 26 (described below). In this case, the input section 22 for receiving the time data is also an integral part of the processor 26. The processor performs a number of different functions, as described below that may be referred to by names of items; in the preferred embodiment of the invention, these items are implemented as software modules.

The memory 28 stores a database 29 of information/data regarding various known appliances. The power consumption of some appliances is variable. For example, a washing machine will consume considerably different amounts of power during different portions of a washing program/cycle and this will differ from program to program. All such data is retained in the memory 28 for each known appliance. The memory 28 may be any suitable computer-readable storage medium, such as a solid-state computer memory, a hard drive, or a removable disc-shaped medium in which information is stored magnetically, optically or magneto-optically. The memory 28, may even be remote from the apparatus and accessible, for example, via a telephone line or over the internet.

The memory 28 may be dynamically updateable, for example by downloading new appliance data. This could be done via the supply wiring 14 itself or, in one optional version, the memory 28 is provided as an IC-card insertable by the user into a slot in the apparatus 20. Manufacturers of appliances provide the necessary appliance data either directly to the consumer, or to the utility company. New IC-cards can be mailed to the user to update their apparatus 20.

The software that the processor 26 runs to perform the analysis may also be stored in the memory 28 and updated as desired in the same ways as the appliance data (e.g. by downloading, by inserting a new medium such as a disc or IC-card, and so on).

The processor 26 receives data from the input section 22, the memory 28 and possibly the clock 24. The processor could be a general purpose processing device or could be a digital signal processor or could be a bespoke hardware device (e.g. FPGA or ASIC) manufactured specifically for implementing one or more embodiments of the invention. -The processor 26 then performs various processing/analysis steps which are described in detail below. Following the is processing/analysis, the processor 26 produces information regarding electrical energy utilisation for some or all of the appliances 12. This information may be transmitted directly to the utility provider. Alternatively, this information may be output by the output section 40 to a user terminal 42 (such as a PC or a dedicated device for utility-use feedback) so that the information can be conveniently presented to the user. The user terminal 42 can be a standard desktop or laptop computer with an attached monitor/display 44 and/or printer 46, or can be a dedicated device.

Although the apparatus 20 and the user terminal 42 are shown as separate devices in Figure 1, they could, of course, be part of the same device.

The output section 40 in the preferred embodiment communicates wirelessly, for example by radio frequencies (RF) link, or optically, or by infrared, or acoustically.

However, it is also possible that the communication with the user terminal 42 is done through the supply wiring 14 if the user terminal 42 is plugged into one of the supply outlets as an appliance. In a further embodiment, the output section 40 can also act as a receiver, such that communication between the apparatus and user terminal 42 is two-way. This enables the user terminal 42 to be used as a further means for updating the appliance data in the memory 28.

The voltage and current values and any other utility values together with the time data are received by the processor 26. From the raw data, the processor calculates a number of coefficients or signature values to characterise the present usage of each utility. Examples of coefficients or suitable signature values for electricity include, but are not limited to: (a) the total real power consumption; (b) the phase difference (angle) between the current and voltage which depends on the load applied by the various appliances 12 and whether it is purely resistive or also reactive, i.e. containing capacitive or inductive loads such as motors and transformers; (c) the root-mean-squared (RMS) current.

Clearly some of the electricity coefficients or signature values mentioned above are averages, typically over a minimum of one cycle of the electricity supply, typically supplied at 50 or 60 hertz so one cycle is approximately 0.02 is seconds. However, mean values of all of the various coefficients or signature values can be calculated over a longer predetermined time interval. The present values of the coefficients or signature values are compared with the running mean value of each coefficient or signature value over the previous cycle or cycles to obtain a change or delta' in each coefficient or signature value.

The processor 26 is shown in more detail in Figure 2. In one embodiment, the processor 26 comprises a signal processing module 50, an event detector 52 comprising one or more detector modules 53A, 53B, 53C, ..., an event processing module 54, an analysis engine 56 comprising one or more analysis modules 57A, 57B, 57C, ..., a fuzzy logic module 58, an event identification module 60 and a correction engine 62.

The signal processing module 50 performs a number of functions and can be implemented in a combination of hardware of software. Some of these are standard such as anti-aliasing filtering and D-A conversion. However, a re-sampling system may also be included for higher accuracy.

-10 -Known event detectors relating to electrical events tend to look for a change in real power and possibly reactive power (e.g. see US 4858141). The present apparatus 20 has an event detector 52 which includes one or more detector modules 53 that are used to detect utility events' relating to the use of one or more appliances 12 (as opposed to events which relate to random noise).

Some event detectors may relate to only one utility, others may relate to combined utility events. The use of multiple detector modules 53 increases sensitivity and reduces the number of false positives. Thus, in a preferred embodiment, the present NIUM apparatus 20 uses a number of detector modules 53 operating in a parallel configuration.

One example of a detector module 53 is a standard electricity event detector module 53A. This is similar to known event detectors where a difference is calculated between the current electrical cycle and the average of the previous n cycles, where a suitable number for n may be 10. The background is averaged in order to reduce the effect of noise' spikes. If the difference is greater than a predetermined threshold, this indicates that an event of interest has occurred.

Advantageously, the threshold is large (e.g. 400VV) in order to avoid noisy loads from triggering events.

Each detector module 53 of the event detector 52 feeds into the event processing module 54 which is configured to analyse the outputs. In a simplified example, each detector module 53 could comprise an output of 0 or 1. In this case, the most logical analysis paradigm is to OR the outputs together since each event detector is designed to detect a very specific feature to fully cover the input space whilst minimising the number of false positives. An alternate methodology would assign a score of between 0 and I from the output of each detector and then combine the results using the operations of Fuzzy Logic or Bayesian Inference.

The core of the NIUM apparatus 20 is contained in the analysis engine 56 which includes functional blocks referred to as analysis modules 57. Each analysis module corresponds to a respective predetermined type of utility usage (e.g. a TRIAC analysis module 57A is concerned with TRIAC-type electricity -11 -usage). Furthermore, each analysis module is arranged to calculate, based on received utility data, a respective confidence value indicative of a confidence that the respective predetermined type of utility usage (e.g. TRIAC-type electricity usage in the example above) has occurred. Many of the analysis modules 57 are non-trivial and act as time domain analysers.

Many of the analysis modules 57 work on high resolution time domain data. At it's core, NIUM is a pattern recognition problem. Pattern recognition techniques often apply standard techniques (e.g. Fourier analysis) to transform the input space into a new space that is more easily analysable and suitable to the problem in hand. Fourier transforms are often used since they allow the extraction of characteristic periodic data that is not visible in time domain signal.

Outside of a few specific examples, the frequency spectrum does not form a useful basis set for the NIUM problem. This can be seen by considering the -specific problem of detecting TRIAC lighting systems in their various modes of operation. Figure 3 shows the Fourier domain representation of the current waveform and it can be seen that, despite the fact that this is all from a single appliance, there is no clear unique signature.

Therefore, in the present NIUM apparatus 20, we focus on creating our own set of features which more enable a more computationally efficient and effective methodology.

The NIUM system described herein uses much higher electrical sampling rates (e.g. in the range of 8 kHz to 80 kHz) than the NILM systems of the prior art so as to extract a more useful set of features than those of the prior art (e.g. US 4858141). By considering only power, it is hard to differentiate between a hypothetical 100W heater and a 100W motor or a 200W lighting system which is at half power. By considering reactive power and harmonic spectra it is possible to gain a little more information; however whilst the harmonic spectrum does contain all of the information contained within the electrical waves, it does not transform the data into a set which is readily understandable.

The NIUM system described herein uses a number of analysis modules 57 which are well suited to identifying the characteristics of particular types of appliances, and thus can vastly improve the accuracy of appliance detection and -12 -energy monitoring. These analysis modules act in a variety of ways and a subset of the analysis modules used are described briefly below.

A TRIAC analysis module 57A uses a technique which can identify the characteristic waveform displayed by a variable brightness lighting system. The s TRIAC analysis module 57A is fully described in UK Patent Application No. 0820812.6 and International Patent Application No. PCTIGB2009/001754, and is briefly summarised here. A TRIAC is a semiconductor device which is used to control the power consumption of resistive devices. A typical TRIAC current waveform is shown in Figure 4. It can be seen that the TRIAC only allows current to pass through the device for part of the cycle, with a sharp edge being present in the current waveform at the switch on/off. When it is non-conducting, the current passed is 0. In the TRIAC analysis module 57A, there is a monitor section, a delta waveform generator and an edge detector. The monitor section is arranged to determine current waveforms comprising sets of values representative of the cyclic waveform of the electric current supply. The delta waveform generator is arranged to calculate the difference between a current waveform and an earlier current waveform, by subtracting the respective sets of values determined by the monitor section, to obtain a delta waveform. The edge detector is arranged to detect an edge or edges in the delta waveform. Using this methodology, the TRIAC analysis module 57A is able to output a confidence that a TRIAC-type event has occurred. Other data may also be output.

A resistive analysis module 57B (as described in UK Patent Application No. 0819763.4 and International Patent Application No. PCT/GB2009/001 754) uses a technique which can identify the characteristic transients developed as a result of the heating of different elements as. found in incandescent light bulbs, space heaters and immersion heating systems. The resistive analysis module 57B is arranged to analyse the received electrical values so as to identify a resistive appliance switch-on event. Having identified a resistive event, the resistive analysis module 576 is further arranged to determine: (i) a first value related to the resistance of the appliance at the time of being switched on; and (ii) a second value related to the resistance of said appliance when operating in a steady state. Using this methodology, the resistive analysis module 57B is able -13 -to output a confidence that a resistive-type event has occurred and parameters relating to that event (e.g. the first and second values mentioned above). Other data may also be output.

An induction motor analysis module 57C (as described in UK Patent Application No. 0913312.5) uses a technique which can identify the characteristic transients found as a result of the acceleration of an induction motor. The induction motor analysis module 57C is arranged to identify the operation of an electrical appliance comprising an induction motor when a path traced by real power values against corresponding reactive power values over a time period of interest comprises one or more substantially circular arcs. Using this methodology, the induction motor analysis module 57C is able to output a confidence that a induction motor-type event has occurred. Other data may also be output. -The detector modules 53 of the event detector 52 are designed to detect events' (e.g. a change in the power consumption of the electricity supply) and are used principally for computational efficiency reasons. Conceptually, it may be possible to run the analysis modules 57 of the analysis engine 56 and the detector modules 53 of the event detector 52 in parallel and combine them at the fuzzy logic module stage with an AND operation (e.g. IF TRIAC-Analysis-Module- Output = high AND Event-Detected = high THEN Conclusion High-Chance-of-TRIAC-Type-Event). However, it is clear that it is more straighiforward to only run the analysis modules 57 of the analysis engine 56 after an event has been detected.

Additionally, some detector modules 53 of the event detector 52 provide an extra analysis module' functionality. For example, the power of a washing machine ramps up over a number of cycles, in comparison to the power ramp of e.g. a kettle which is effectively instantaneous. Thus a detector module 53 can also provide analysis module functionality: It will be understood that such a functional block exists in both the event detector 52 and also the analysis engine 56 and that it is an implementation issue as to whether there are one or two physical modules.

-14 -Around twenty analysis modules 57 are presently used in the NIUM apparatus 20, and thus a system is required to analyse the outputs and draw meaningful conclusions, whilst remaining robust in the high noise environment. A problem facing the NILM (or NIUM) designer is that there are a huge number of appliances in existence and thus, for a single analysis module, it is not straightforward to produce a robust mathematical model for e.g. the level of 3rd harmonic in a fridge motor. This problem is made worse by the large amounts of effectively random noise which is superimposed over the signal. A further problem is that many appliances can be viewed as a combination of other smaller functional blocks. For example, whilst vacuum cleaner consists of a simple induction motor, a tumble dryer has both a heater and a tumble dryer and thus on occasion may display characteristics of both a motor and a heater. The fuzzy logic module 58 has been adopted to provide a language and structure to deal with these fundamentally vague concepts.

Each analysis module works internally in a different numeric range. For example, the induction motor analysis module 57C indicates the presence of an induction motor if one of the output values is less than 5e-3. In contrast, the resistive analysis module 57B may indicate a strong match if the output is greater than 0.99. However, for a fuzzy system to operate, it is necessary to fuzzily' these inputs. The fuzzification process is effectively a non-linear transform, and is known as the membership function. The membership function can take any form. As an example, the resistive analysis module 57B may have a membership function as shown in Figure 5. Thusit can be seen that for resistive inputs of less than 0.99, the Degree Of Membership (DOM) drops off very rapidly.

The goal of a Fuzzy Inference System (F IS) is to attempt to infer conclusions based on uncertain inputs. For example, the goal having detected an event is to ascertain whether that event was indicative of a shower turning on, without necessarily having seen that model of shower before. From a human reasoning point of view, one may attempt to detect the shower based on a number of rules: -15 -IF Event-Power = high AND Motor-Power low AND Resistive-Analysis-Module-Output high

THEN Conclusion Shower-Event

On account of noise and uncertainty, we have loose definitions for high' and low' in each case. Such a problem can be handled well by a Fuzzy Inference System, such as the fuzzy logic module 58 which is described in more detail below with reference to Figure 6.

For clarity, Figure 6 only shows a small number of inputs and outputs.

Such a structure is inspired by ANFIS: Adaptive-Network-Based Fuzzy Inference System' by Jyh-Shing Roger Jang (IEEE Transactions on Systems, Man and Cybernetics, Vol. 23, No. 3, May/June 1993). Such an architecture is referred to as a Sugeno Fuzzy Inference System and the output is indicative of the conclusion of the system -e.g. a cooker has been turned on.

In Figure 6, El and E2 represent the outputs of two of the analysis modules 57. E100M1 and E1DOM2 represent the membership functions for the first analysis module, and E200M1 and E200M2 represent the membership functions for the second analysis module. Multiple DOMs for each analysis module 57 are used to allow the fuzzification over the full range of analysis module outputs. For example, Consider the use of two fuzzy logic rules: IF Analysis-Module-i-output = large AND Analysis-Module-2-output = large THEN Conclusion-i IF Analysis-Module-i-output medium AND Analysis-Module-2-output = large THEN Conclusion-2 In this case, it is clear that Analysis-Module-I needs to be connected to two membership functions -one to define large' and one to define medium.' The 1-Norm stage represents a fuzzy AND' operator. This can be implemented in a number of ways, though the common methods are either multiplication, or a MIN' operator of the inputs.

The Normalisation layer calculates the ratio of each rule's firing strength to the sum of all the rules' firing strengths and outputs a normalised rule strength.

The firing strength' of a rule may be thought of as an output level for each rule, i.e. the strength of each rule (see also the ANFIS: Adaptive-Network-Based Fuzzy Inference System' article by Jyh-Shing Roger Jang).

-16 -The rule output stage combines the output function of the T-Norm stage with the rule strength. In our simple example (referred to as a Type 1' system in the ANFIS: Adaptive-Network-Based Fuzzy inference System' article by Jyh-Shing Roger Jang), then our output function is a constant such that the rule output stage will output a set of rules and weights/confidences -e.g. Cooker 0.8, Hoover 0.2, where Cooker' and Hoover' are the output functions of the T-Norm stage and 0.8' and 0.2' are the associated rule weights.

The final output stage aggregates all of the rules and produces a single output. In a Type 3' system where each rule output is a numeric value, this stage acts as a summation of all incoming signals. In our example, the output is simply an amalgamation (e.g. an event has the characteristics that make it belong to the cooker set with 0.8 membership, and the hoover class with 0.2 membership).

In conclusion, such a system therefore leads to the designer being able to implement a rule set including rules such as: IF Analysis-module-i -output strong AND Analysis-module-2-output weak OR Analysis-module-i-output weak 2 0 AND Analysis-module-3-output strong Thus, one can combine the outputs of the analysis modules 57 in a logical fashion whilst accounting for the inherent vagueness which defines the process.

Following our noisy measurement of an unknown appliance, we attempt to classify how cooker like' that appliance is compared to how similar it is to other possible appliances by using a set of basic rules.

To summarise, each analysis module 57 effectively provides an outputs which is indicative of a confidence that the event detected by the event detector 52 and event processing module 54 corresponds to a particular predetermined type of utility usage which is the subject of that analysis module 57. For example, the TRIAC analysis module 57A provides an output indicative of a confidence -17 -that a TRIAC-type event has occurred. The output numbers from each of the analysis modules 57 could be combined using the laws of Boolean logic (by constraining the values to 0, 1) or Bayesian logic. However, in the present system, the outputs are advantageously combined using fuzzy logic in the fuzzy logic module 58. One or more of the outputs of the analysis modules 57 are combined using the rules of fuzzy logic in the fuzzy logic module 58 to classify the event. One or more of the outputs of the analysis modules 57 provide information to help match the event.

Each analysis module 57 outputs a respective confidence value indicative of a confidence that the associated predetermined type of utility usage has occurred. Fuzziness is a useful principle since it is not possible to derive meaningful probabilities for many of the events observed and, when combining large numbers of analysis module outputs using the laws of probability, we very quickly derive an answer which is mathematically meaningless, though with the danger that it is perceived to be a precise probabilistic answer.

The outputs of the fuzzy logic module 58 include a number of fuzzy confidences in classification of an event along with various parameters relating to the event itself. An example output of the fuzzy logic module 58 is shown in

Table 1.

Analysis module Confidence Parameter I Parameter 2 Resistive 0.9 500W 100W Induction motor 0.3 1000W 0.23 TRIAC 0.02 100 0.1

Table I

Following this exemplary event, we can see that our confidence is high that this is a resistive event and has two parameters (in this case the peak power and the steady state power). Our confidence that it was an induction motor is low, but not insignificant. The two parameters in this case would represent different -18 -values. The fuzzy logic module 58 concludes that it is very unlikely that the event was a TRIAC-type event.

The NIUM system can be parameterised in a number of ways. Each analysis module 57 can be parameterised (for example, in our trivial example, our induction motor analysis module 57C runs 50 cycles after an event is detected.

However, it may be better to run 40 cycles later). Further, each membership function can be parameterised. Thus, with a suitable training set, it is possible to provide an automated system that performs off-line learning of the most suitable parameters. Such a system is referred to as an ANFIS -Adaptive-Network-based Fuzzy Inference System. Such systems can perform very well since they allow a highly non-linear mapping from the input to output state (a characteristic shared with neural networks). However, in contrast to neural networks, the underlying architecture of the present system (embodied in the fuzzy logic module 58 of the processor 26) is simple to understand thus it can be easily designed and maintained. This is in contrast to many neural network implementations which very much operate as a black box' type system where good results can be obtained at the expense of highly limited visibility of the actual reasoning process.

Following the successful detection and classification of the event, the event identification module 60 is used to identify the specific appliance which is the source of the event. At this point, we compare the characterising parameters of the event to those of known appliances' held in the database 29 of the memory 28 and look for suitable matches. For example, considering our event above from Table 1, we would look to fuzzily match for resistive type appliances which match the identified parameters. If there is no good match, we would add a new appliance. However, if there was a high chance that the event was also an induction motor then we would look to match for an induction motor as well, though as a rule, the goal of the fuzzy logic module 58 is to produce only one clear candidate for matching.

-19 -The correction engine 62 acts in parallel to the main system and continually analyses the database 29 to look for inconsistencies in the matching.

An initial problem in matching is how to set the matching tolerances since there is no prior measure of the variability of the parameters to be matched. For example, a light bulb will have a measured power consumption which varies by only 1% plus background noise. However, the power consumption of a vacuum cleaner may vary by as much as 5%, hence it is a non-trivial problem to decide whether on the edge of tolerance, one should create a new appliance, or match to an existing appliance. The correction engine is designed to cope with such io problems and to correct any incorrect appliance identifications.

Although preferred embodiments of the invention have been described, it is to be understood that these are by way of example only and that various modifications may be contemplated.

Claims

-20 -CLAIMS: 1. A method of non-intrusive utility monitoring for monitoring the use of at least one utility supplied to a plurality of appliances, the method comprising: receiving utility values representative of the total use of the at least one utility by the plurality of appliances; analysing the received utility values using a plurality of analysis modules, wherein each analysis module corresponds to a respective predetermined type of utility usage, and wherein each analysis module is arranged to calculate, based on the received utility values, a respective confidence value indicative of a confidence that the respective predetermined type of utility usage has occurred; and performing a fuzzy logic analysis of thecalculated confidence values so as to identify the operation of an appliance.
2. The method of claim I wherein the predetermined type of utility usage for one of the plurality of analysis modules is usage representative of a resistive device with a relatively constant steady-state load.
3. The method of claim I or claim 2 wherein the predetermined type of utility usage for one of the plurality of analysis modules is usage representative of a predominantly resistive device employing intra-cycle switching to variably control the power supplied to a load.
4. The method of any preceding claim wherein the predetermined type of utility usage for one of the plurality of analysis modules is usage representative of an induction motor wherein a path traced by real power values against corresponding reactive power values over a time period of interest comprises one or more substantially circular arcs.
5. The method of any preceding claim wherein the at least one utility comprises electricity, and the utility values comprise values representative of the -21 -electrical current and/or the electrical voltage supplied to the plurality of appliances.
6. The method of any preceding claim wherein the at least one utility comprises one or more of water, gas and oil.
7. The method of any preceding claim further comprising detecting at least one utility event based on the received utility values.
8. The method of any preceding claim wherein each confidence value is a degree of membership of a respective membership function corresponding to the respective predetermined type of utility usage.
9. A computer program comprising computer-executable code that when executed on a computer system, causes the computer system to perform a method according to any preceding claim.
10. A computer-readable medium storing a computer program according to claim 9.
11. A computer program product comprising a signal comprising a computer program according to claim 9.
12. A non-intrusive utility monitoring apparatus for monitoring the use of at least one utility supplied to a plurality of appliances, the apparatus comprising: an input section arranged to receive utility values representative of the total use of the at least one utility by the plurality of appliances; a plurality of analysis modules, wherein each analysis module corresponds to a respective type of utility usage, and wherein each analysis module is arranged to analyse the received utility values so as to calculate, based on the received utility values, a respective confidence value indicative of a confidence that the respective type of utility usage has occurred; and -22 -a fuzzy logic module arranged to perform a fuzzy logic analysis of the calculated confidence va'ues so as to identify the operation of an appliance.
13. The apparatus of claim 12 further comprising a processor which comprises s the plurality of analysis modules and the fuzzy logic module.
14. A non-intrusive utility monitoring apparatus substantially as herein described with reference to Figures 1, 2 and 6 of the accompanying drawings.
15. A method of non-intrusive utility monitoring, the method being substantially as herein described with reference to Figures 1, 2 and 6 of the accompanying drawings.Amendment to the claims have been filed as follows CLAIMS: 1. A method of non-intrusive utility monitoring for monitoring the use of at least one utility supplied to a plurality of appliances, the method comprising: receiving utility values representative of the total use of the at least one utility by the plurality of appliances; analysing the received utility values using a plurality of analysis modules, wherein each analysis module corresponds to a respective predetermined type of utility usage, and wherein each analysis module is arranged to calculate, based on the received utility values, a respective confidence value indicative of a confidence that the respective predetermined type of utility usage has occurred; and performing a fuzzy logic analysis of the calculated confidence values so as to identify whether one or more of the predetermined types of utility usage has occurred.2. The method of claim 1 wherein each analysis module is further arranged to calculate, based on the received utility values, one or more respective characterising parameters associated with the respective predetermined type of utility usage, and wherein the method further comprises identifying the operation of a specific appliance based on the fuzzy logic analysis and the calculated * characterising parameters. ** ** * . S* *:. 3. The method of any preceding claim wherein the predetermined type of utility usage for one of the plurality of analysis modules is usage representative of a resistive device with a relatively constant steady-state load.4. The method of any preceding claim wherein the predetermined type of utility usage for one of the plurality of analysis modules is usage representative of a predominantly resistive device employing intra-cycle switching to variably control the power supplied to a load.5. The method of any preceding claim wherein the predetermined type of utility usage for one of the plurality of analysis modules is usage representative of an induction motor wherein a path traced by real power values against corresponding reactive power values over a time period of interest comprises one or more substantially circular arcs.6. The method of any preceding claim wherein the at least one utility comprises electricity, and the utility values comprise values representative of the electrical current and/or the electrical voltage supplied to the plurality of appl lances.7. The method of any preceding claim wherein the at least one utility comprises one or more of water, gas and oil.8. The method of any preceding claim further comprising detecting at least one utility event based on the received utility values.9. The method of any preceding claim wherein each confidence value is a degree of membership of a respective membership function corresponding to the respective predetermined type of utility usage. * 010. A computer program comprising computer-executable code that when executed on a computer system, causes the computer system to perform a method according to any preceding claim. * *11. A computer-readable medium storing a computer program according to * claim 10. *12 A computer program product comprising a signal comprising a computer program according to claim 10.13. A non-intrusive utility monitoring apparatus for monitoring the use of at least one utility supplied to a plurality of appliances, the apparatus comprising:an input section arranged to receive utility values representative of the total use of the at least one utility by the plurality of appliances; a plurality of analysis modules, wherein each analysis module corresponds to a respective type of utility usage, and wherein each analysis module is arranged to analyse the received utility values so as to calculate, based on the received utility values, a respective confidence value indicative of a confidence that the respective type of utility usage has occurred; and a fuzzy logic module arranged to perform a fuzzy logic analysis of the calculated confidence values so as to identify whether one or more of the predetermined types of utility usage has occurred.14. The apparatus of claim 12 further comprising a processor which comprises the plurality of analysis modules and the fuzzy logic module.15. The apparatus of claim 13 wherein each analysis module is further arranged to calculate, based on the received utility values, one or more respective characterising parameters associated with the respective predetermined type of utility usage, and wherein the apparatus further comprises an event identification module arranged to identify the operation of a specific appliance based on the fuzzy logic analysis and the calculated characterising *..parameters.* ***.* * S
16. The apparatus of claim 15 further comprising a processor which comprises the plurality of analysis modules, the fuzzy logic module and the event identification module. 25
17. A non-intrusive utility monitoring apparatus substantially as herein described with reference to Figures 1, 2 and 6 of the accompanying drawings.
18. A method of non-intrusive utility monitoring, the method being substantially as herein described with reference to Figures 1, 2 and 6 of the accompanying drawings.